Sensor calibration

FSR Sensor general information

The tactile skin on the robotic gripper is composed of two distinct layers. The external layer is fabricated using multi-material 3D printing techniques, utilizing a rubbery polymer known as Agilus30. The internal layer provides core sensing abilities and is made from a flexible sensor array that mimics human tactile perception through a dense array of mechanoreceptors.

The RX-MI0404S model represents a state-of-the-art distributed flexible thin-film pressure sensor, engineered via advanced precision printing methods. This process involves the deposition of nanoscale piezoresistive materials and conductive silver paste onto a supple substrate, followed by a drying and curing phase to solidify the sensor.

The sensor operates on the principle that its electrical resistance inversely correlates with the applied pressure. This correlation manifests as a power function between resistance and pressure, while the reciprocal of resistance maintains an approximate linear relationship with pressure. Each sensing element within the array functions as a pressure-variable resistor.

The depicted sensor comprises a tactile array with a 10 mm by 10 mm sensing area, arranged in a 4x4 grid. This grid consists of 16 individual sensing units, each with dimensions of 2 mm by 2 mm, culminating in a actual sensing area of 64 mm².

Although the resistance is linear proportional its pressure, to use the the sensor, careful calibration and evaluation is required.

Sensor Evaluation Metrics

Dynamic Range

Definition: The dynamic range assesses the sensor's capability to detect forces across a broad spectrum, from very light to significantly heavy. This range is crucial for mimicking the versatility of human skin.
Relevance: Essential for applications requiring interaction with objects of varying weights and for ensuring sensitive and adaptable force detection.

Accuracy

Definition: Accuracy refers to the sensor's precision in reporting applied forces, validated by comparing readings against known standards.
Relevance: Critical for ensuring the reliability of force measurements and for tasks demanding high precision in force application.

Spatial Resolution

Definition: Spatial resolution tests the sensor's ability to differentiate between closely spaced stimuli.
Relevance: Mirrors the high spatial resolution of human tactile perception, important for detailed texture and shape recognition.

Temporal Resolution

Definition: Temporal resolution measures the sensor's speed in detecting and responding to changes in force, capturing the dynamics of force variations.
Relevance: Key for simulating the rapid response of human skin to different stimuli, enabling the sensor to track and respond to quick changes in force.

Sensitivity

Definition: Sensitivity evaluates the sensor's responsiveness to slight changes in force.
Relevance: Reflects the high sensitivity of human skin, necessary for detecting delicate touches and minor pressure differences.

Drifting

Definition: The degree of dispersion between the measurements and actual values in repeated experiments. Drifting describes the sensor's baseline or reference point shift over time, without changes in external conditions, leading to signal degradation and measurement inaccuracies.
Relevance: Affects the sensor's accuracy and reliability, highlighting the need for regular calibration or algorithms to correct or compensate for these shifts.

Experiments setup

As depicted in the left figure, the calibration experiment employed the following four components:

Weights ranging from 10 to 200 grams.
The FSR sensor, model rxm0404s, which required calibration and testing
A small platform produced through 3D printing. The platform's smaller surface area coincides with the receptive face of the FSR sensor, while its larger surface is designated for bearing loads.
The M3813B 6-axis load cell, model M3813B , continuously monitors the load force in the Z-axis direction.

Sensor calibration

Since the sensor readings are directly proportional to the applied pressure, calibration can be achieved by multiplying the raw readings by a specific number to equate them with the gravitational force produced by the weights. However, it's important to note the FSR sensor's limited receptive area. Utilizing the previously mentioned small platform is crucial to maximize the congruence of their contact surfaces. Despite this, there may still be variations in sensor readings, with some points having valid values while others do not. Thus, when converting pressure to force, one should not simply multiply by the sensor's total surface area of 100 mm^2. Instead, it's necessary to first identify which points are registering readings. Based on this, calculate the correct contact area, which is 6.25 mm^2 per pixel, and then multiply by the pressure to obtain the accurate contact force. Finally, summing these correct contact forces and calculating their ratio to the force exerted by the weights allows for the precise calibration of the sensor.

Sensor Evaluation

After the calibration of the FSR sensor, it was fixed onto the 6-axis load cell. Subsequently, the platform was positioned on the FSR sensor. The platform was subjected to a continuous load, and the reading of the FSR sensor and the load cell were recorded at the same time. This procedure was repeated five times, and all the outcomes are illustrated in the figure below, where its X axis is the reading from the load cell, the Y axis is the reading from the FSR sensor.

Results

Since the Load cell has been tested to have a extremely high accuracy regarding to the force reading. So it would be used to act as as ground truth equals to generated force.

As for the graph of FSR reading VS the Generated force. The perfect results should be Y=X, but not all data range follows this trend.

The data can be approximately divided into three segments, represented by blue, yellow, and green, corresponding to three states of the sensor.

In the blue range (0-2N), a significant portion of the FSR readings were below the true value, with high randomness in data points, indicating an inability to obtain stable readings in this range. The sensor's minimum detection threshold is identified as 2N. In the yellow range (2-25N), the data were highly concentrated and stable, aligning closely with the actual values, signifying this as the sensor's operational range. In the green range, beyond 25N, the sensor showed high dispersion, suggesting the upper limit of the operational range is 25N.

The analysis of the sensor's performance within the 2-25N force range reveals notable efficiency. The red trend line, representing this range, is based on a linear regression formula: y = 1.01x - 0.44. This equation highlights a sensitivity of 1.01 and a minimal offset of 0.34, underlining the sensor's exceptional accuracy in this range. Moreover, the standard deviation within this range is a mere 0.11, emphasizing its consistent performance. The sensor also demonstrates a relative accuracy of 6.59% in the 2-25N range.

Contrasting with its performance in the 2-25N range, the sensor exhibits variable accuracy in other force ranges. Notably, the Relative Accuracy escalates to 63.13% for forces between 0-2N, while it decreases to 10.18% for forces exceeding 25N. This variation is further highlighted by the standard deviation values, which are substantially different from those in the 2-25N range. Specifically, the standard deviation is 1.23 for forces between 0-2N and 0.13 for forces beyond 25N.

Additionally, the linear regression analysis provides further insights into the sensor's performance outside the 2-25N range. For forces between 0-2N, the trend line equation is y = 0.54x + 0.64. For forces beyond 25N, the equation is y = 0.51x + 16.79. These equations indicate a marked decrease in sensitivity (approximately 0.5) and larger offsets in both ranges, signifying a notable decrease in accuracy for sensor work outside the designed force range.

sd1689426887_2.mp4

Page updated

Google Sites

Report abuse